A new parallel genetic algorithm

Author

Tan, Ling ; Taniar, David ; Smith, Kate A.

Author_Institution

Sch. of Bus. Syst., Monash Univ., Clayton, Vic., Australia

fYear

2002

fDate

6/24/1905 12:00:00 AM

Firstpage

284

Lastpage

289

Abstract

One problem of propagating the globally fittest individual via neighbourhood evolution in both the island model and the cellular model of existing parallel genetic algorithms (PGAs) is that the migration of the globally best individual is delayed to non-adjacent processors. This may cause an inferior search in those sub-populations. The propagation delay of the globally best individual is proportional to the network distance between two processors. Delayed migration of the best individual in PGAs is an essential deviation from the sequential version of the genetic algorithm, in which the best individuals are always used to compete with other individuals. To solve this problem, this paper proposes an extended version of the island PGA called the Virtual Community PGA (VC-PGA). The VC-PGA is applied in a case study of optimizing the parameters of a backpropagation neural network classifier

Keywords

backpropagation; delays; genetic algorithms; neural nets; parallel algorithms; pattern classification; Virtual Community parallel genetic algorithm; backpropagation neural network classifier; case study; cellular model; globally fittest individual propagation; inferior sub-population search; island model; neighbourhood evolution; nonadjacent processors; parameter optimization; processor network distance; propagation delay; Australia; Electronics packaging; Genetic algorithms; Neural networks; Propagation delay; Read only memory; Virtual colonoscopy;

fLanguage

English

Publisher

ieee

Conference_Titel

Parallel Architectures, Algorithms and Networks, 2002. I-SPAN '02. Proceedings. International Symposium on

Conference_Location

Makati City, Metro Manila

ISSN

1087-4089

Print_ISBN

0-7695-1579-7

Type

conf

DOI

10.1109/ISPAN.2002.1004301

Filename

1004301